This repository is a curated list of research papers focused on "Inference/Test Time Scaling/Computing" for Large Language Models (LLMs). It aims to provide a comprehensive resource for researchers and practitioners interested in improving LLM performance by allocating additional computational resources during inference, rather than solely relying on training-time scaling.

How It Works

The project collects and categorizes academic papers that explore various techniques for inference-time scaling. These techniques often involve methods like generating multiple reasoning paths (e.g., Chain-of-Thought, Tree-of-Thoughts), using verification mechanisms (e.g., reward models, self-correction), or employing search algorithms (e.g., Monte Carlo Tree Search) to refine outputs and improve accuracy on complex tasks. The goal is to understand how to optimize the trade-off between computational cost and performance gains at inference time.

Quick Start & Requirements

• Installation: This is a paper list, not a runnable software package. No installation is required.
• Requirements: Access to academic paper repositories (e.g., arXiv) is needed to view the listed research.
• Resources: No computational resources are required to browse the list.

Highlighted Details

• Comprehensive collection of recent research on inference-time scaling.
• Categorization of papers by technique and application area.
• Includes links to arXiv PDFs for easy access to research papers.
• Facilitates understanding of emerging trends in LLM reasoning and optimization.

Maintenance & Community

The project is actively maintained and welcomes contributions from the community. Users can contribute by opening pull requests or issues to suggest missing papers or corrections.

Licensing & Compatibility

The repository itself is likely under a permissive license (e.g., MIT, Apache 2.0) as it is a collection of links to research papers. The individual papers retain their original publication licenses.

Limitations & Caveats

This is a curated list of papers and does not provide any executable code or models. The effectiveness of the techniques described in the papers may vary depending on the specific LLM, task, and implementation details.